The 2024 Google Core Update and AI Sites: What Got Deindexed and Why

In March 2024, something happened that a lot of AI publishers had quietly assumed never would. Entire sites — not a page here, a ranking there, but whole sites — vanished from Google’s index overnight. Some of them had been printing thousands of AI-generated articles a month. This is the story of what actually got removed, and the more useful question of why.

Key takeaways

• The March 2024 core update was unusually large and ran roughly six weeks (March 5 to April 19, 2024), the longest core update Google had run to that point.
• Google didn’t deindex sites for *using* AI. It deindexed sites that mass-produced thin, unhelpful pages — and many of those happened to be AI-built.
• Manual actions hit hardest. Some sites had every page dropped from the index, not just demoted in rankings.
• The clearest pattern was scale plus sameness: hundreds or thousands of near-identical pages with no firsthand value.
• AI-assisted sites that added real expertise and editing mostly survived, which tells you the tool was never the trigger.

What actually happened in March 2024

Google announced the March 2024 core update on March 5, alongside a batch of new spam policies. The core update itself didn’t finish rolling out until April 19 — about six weeks, which made it the longest core update Google had run at the time. Most core updates wrap in one or two weeks. This one ground on. Google also described it as its most complex core update yet, and used it to fold the old standalone helpful content system into its core ranking.

Two separate things were happening at once, and people kept confusing them.

The first was the core update: an algorithmic reassessment of content quality across the whole web. Core updates don’t punish individual sites so much as recalibrate what Google considers helpful. When the dust settles, some sites rank lower, some rank higher, and nobody gets a notice in their inbox.

The second was a wave of manual actions tied to the new spam policies. These were targeted. A human or a system flagged specific sites, and Google sent a notification through Search Console. The harshest of these didn’t just demote pages. They removed them from the index entirely. That’s the deindexing everyone was talking about.

So when you read that “AI sites got deindexed,” the precise version is: a number of sites received manual actions for spammy, mass-produced content, and many of those sites happened to be running on AI.

What got deindexed, specifically

In the days after the spam policies kicked in, SEO researchers started cataloguing the sites that disappeared. A pattern showed up fast, and it had almost nothing to do with whether a model wrote the words.

The deindexed sites tended to share these traits:

• Enormous page counts built quickly. Sites that had gone from a few dozen pages to thousands in a matter of weeks or months.
• Templated sameness. Page after page following an identical skeleton — same headings, same structure, only the topic keyword swapped.
• No firsthand anything. No original photos, no tests, no data the site collected itself, no author who’d clearly done the thing they were writing about.
• Pure keyword targeting. Each page existed to capture one search query and nothing else.

Some of these were openly AI-first projects whose owners had written about their methods publicly, which made them easy to study after the fact. The lesson wasn’t subtle. The sites weren’t removed because a detector sniffed out GPT. They were removed because the *strategy* was scaled content abuse — producing volume to game rankings rather than to help anyone.

A mini-scenario that makes the line clear

Picture two sites about houseplants.

Site A ran a script across a spreadsheet of 1,200 plant names and generated a “How to care for [plant]” page for each. Same eight headings every time. Watering, light, soil, repotting, and so on. No photos the owner took, no mention of a plant they actually killed, nothing a reader couldn’t get from the first result already ranking. Twelve hundred pages, one weekend.

Site B used AI to draft articles too, but the owner is a person who actually grows these plants. They added their own photos, notes on what went wrong in their apartment’s bad north light, and a few opinions that contradict the usual advice. Forty pages over a year. Each one earns its place.

Both used AI. After March 2024, Site A was gone and Site B was fine. The model didn’t decide that. The intent and the value did.

Why Google could afford to do this

For a long time, the friction of writing protected the web from a flood of junk. Producing a thousand decent pages took real time, so most people didn’t bother. Generative AI removed that friction almost completely. Suddenly a single person could publish at the scale of a content farm, and many did.

Google noticed the volume problem and updated its language to match. The old spam rule was called “automatically-generated content,” and it leaned on the idea that machine-made text was inherently the issue. People dodged it by having a human lightly touch the output. The 2024 version of the policy dropped that framing entirely — it now applies “no matter how it’s created,” whether by automation, humans, or a mix. That closed the loophole and moved the question from *”did a machine make this?”* to *”does this page deserve to exist?”*

It helps to understand the mechanics underneath all this. Quality systems and AI detectors both lean on statistical signals — how predictable and how varied your sentences are. We unpacked the statistics behind how detectors read your text if you want the details. Google’s ranking systems aren’t the same as an AI detector, but they reward the same underlying thing: writing with real substance and rhythm instead of flat, templated filler.

Why some AI content survived just fine

Here’s the part the panic headlines left out. Plenty of sites using AI heavily did *not* get hit. Some kept ranking and a few gained.

What separated the survivors wasn’t a secret. It was the boring stuff:

• A reason for each page to exist beyond “it ranks for a keyword.”
• Original input — a test, a number, an opinion, a photo, a customer story.
• Actual editing, where a person cut filler and fixed claims.
• A pace the team could stand behind, instead of a firehose nobody reviewed.

If you take one thing from the 2024 fallout, take this: Google never penalized the tool. It penalized the absence of a human who cared. AI made it trivially easy to publish content nobody had bothered to make good, and Google adjusted to deal with that.

How to stay on the safe side now

You don’t have to swear off AI. You have to use it like a drafting assistant, not a printing press.

Have a reason for every page. Before you publish, answer honestly: who is this for, and what do they get here that they can’t get elsewhere? If the only answer is “it ranks,” don’t make it.

Add something only you have. Google calls this information gain. It’s the gap between unoriginal and useful.

Edit like you mean it. Cut the filler, fix the facts, break up the relentlessly even paragraphs that scream template. A surface-level rewrite of a worthless page leaves you with a worthless page that reads slightly differently.

Don’t scale faster than you can vouch for. If you can’t stand behind 50 pages this month, don’t ship 500.

A quick pass to check whether a draft still reads machine-flat can help you spot the parts that need a human’s attention — not to trick anyone, but because flat, predictable writing is the same signal that flagged thin work to Google in the first place. If you’re curious about the wider picture, you can browse the rest of our writing on AI and search.

Frequently asked questions

Did Google deindex sites just for using AI in 2024? No. There’s no evidence Google removed pages because a model helped write them. What got hit was the strategy behind many AI sites — pumping out thin, unoriginal pages to catch traffic. The AI was the method, not the offense.

What’s the difference between deindexing and a ranking drop? A ranking drop means your pages stay in the index but show up lower. Deindexing removes them entirely. In 2024, core update effects mostly demoted sites; separate manual actions deindexed some outright, and those came with a Search Console notice.

When exactly did the March 2024 core update happen? It was announced March 5 and finished rolling out April 19, 2024 — about six weeks. The spam policies announced the same day, including scaled content abuse, ran on a faster timeline that wrapped around March 20.

Can a deindexed site recover? Sometimes, slowly. For a manual action, fix the problem, remove or genuinely improve the thin pages, and file a reconsideration request. For algorithmic effects, there’s no request — you improve quality and wait for a future update. Surface rewrites of junk don’t restore trust.

Is it safe to use AI for content now? Yes, used well. The line is about value and intent, not the tool. An AI-assisted article with original research and editing is fine. A thousand templated pages overnight is the risk.

The honest summary

The 2024 deindexing wave wasn’t Google declaring war on AI. It was Google catching up to a sudden flood of pages that existed for search engines instead of people. The fix has always been the same: make each page something a reader is glad they found. Want your drafts to read like a person actually cared before you publish? Run a draft through PaperBleach, and if you’re working at any real volume, see how the plans compare.